Abstract: For timevarying feature of industrial process, the algorithm called adaptive moving windows of PCA(principal component analysis) could update system model by collecting data, which can improve modeling precision and diagnostic accuracy. This algorithm is based on two assumptions: (1) The data used for updating model are collected during process; (2) Time series were independent for the collected data. As this algorithm could not recognize whether the data are collected from stable process or fault process. For overcoming this problem, this paper proposed relative variable quantity to identify the two states. Industry practice showed that most industrial processes exist time series dependent, so, dynamic PCA(DPCA) should be considered. This paper proposed the key parameter of timelag parameter for DPCA and used the parameter to compute and improve the moving windows algorithm. Finally, by simulation test, the effectiveness of the new algorithm was verified.